Coverage Problem 2

7/27/2019 Coverage Problem 2

1/15

78 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART A: SYSTEMS AND HUMANS, VOL. 35, NO. 1, JANUARY 2005

Energy-Efficient Deployment of IntelligentMobile Sensor NetworksNojeong Heo and Pramod K. Varshney, Fellow, IEEE

AbstractMany visions of the future include people immersedin an environment surrounded by sensors and intelligent devices,which use smart infrastructures to improve the quality of life andsafety in emergency situations. Ubiquitous communication enablesthese sensors or intelligent devices to communicate with each otherand the user or a decision maker by means of ad hoc wireless net-working. Organization and optimization of network resources areessential to provide ubiquitous communication for a longer du-ration in large-scale networks and are helpful to migrate intelli-gence from higher and remote levels to lower and local levels. Inthis paper, distributed energy-efficient deployment algorithms formobile sensors and intelligent devices that form an Ambient In-

telligent network are proposed. These algorithms employ a syner-gistic combination of cluster structuring and a peer-to-peer deploy-ment scheme. An energy-efficient deployment algorithm based onVoronoi diagrams is also proposed here. Performance of our algo-rithms is evaluated in terms of coverage, uniformity, and time anddistance traveled until the algorithm converges. Our algorithmsare shown to exhibit excellent performance.

Index TermsAmbient intelligence, deployment, distributedalgorithms, energy-efficiency, mobile wireless networks, wirelesssensor networks (WSN).

I. INTRODUCTION

DESIGN and deployment of infrastructured networks,

such as a cellular network, has matured over the last twodecades. In such networks, mobile users access the network via

fixed base stations. Planning and deployment of these networks

is carried out based on radio propagation and terrain models,

with the goal of maximizing radio coverage. More recently,

there has been a great deal of interest in ad hoc networks.

These networks employ fixed or mobile nodes and dynami-

cally organize themselves into a network without requiring an

infrastructure. In ad hoc networks, each node acts not only as

an end node, but also as a router. One important aspect in the

design of these networks is the initialization procedure and

establishment of the routing structure. In these networks, a new

paradigm is considered, where power usage instead of band-width is of primary concern. Extending system lifetime and

robustness to unpredictable dynamics, rather than optimizing

channel throughput or minimizing the number of nodes, is the

biggest challenge during the design of these networks. Most

research on ad hoc networks has focused on issues such as the

Manuscript received October 20, 2003; revised March 9, 2004 and June 21,2004. This paper was recommended by Guest Editor G. L. Foresti.

The authors are with the Department of Electrical Engineering and Com-puter Science, Syracuse University, Syracuse, NY 13244 USA (e-mail:[email protected]; [email protected]; [email protected]).

Digital Object Identifier 10.1109/TSMCA.2004.838486

development of routing protocols and quality of service and not

on topology and deployment.

Wireless sensor networks (WSN) that employ ad hoc net-

working have become an area of intense research activity. This

is due to the availability of inexpensive sensors for sensing and

control and technical advances in sensors, wireless communica-

tions and networking, and signal processing. Many applications

are envisaged including: environment and habitat monitoring;

wild fire detection; inventory tracking; biomedical analysis; per-

vasive computing; battlefield surveillance; and urban search-

and-rescue operations, especially in hazardous situations. WSNoperate under limited radio coverage and attempt to conserve

bandwidth and battery power. Much research on this issue is

underway ranging from the development of power-saving hard-

ware [18] to power-efficient medium access control (MAC) and

routing protocols [11], [27], and to the development of col-

laborative signal processing and power-aware algorithms [14].

Sensor nodes are generally assumed to be fixed and randomly

placed. The number of sensors is assumed to be quite large so

that coverage of the surveillance area is not an issue. Not much

attention has been paid to optimization in terms of the number

of nodes or their topology.

One of the key issues in this area is the deployment of mobile

sensor nodes in the region of interest (ROI), where interestingevents might happen and the corresponding detection mecha-

nism is required. Before a sensor can provide useful data to

the system, it must be deployed in a location that is contextu-

ally appropriate. Optimum placement of sensors results in the

maximum possible utilization of the available sensors [23]. The

proper choice for sensor locations based on application require-

ments is difficult. The deployment of a static network is often ei-

ther human monitored or random. Though many scenarios adopt

random deployment for practical reasons such as deployment

cost and time, random deployment may not provide a uniform

sensor distribution over the ROI, which is considered to be a

desirable distribution in mobile sensor networks. Uneven nodetopology may lead to a short system lifetime. Self-deployment

methods using mobile nodes [10], [25], [28] have been proposed

to enhance network coverage and to extend the system lifetime

via configuration of uniformly distributed node topologies from

random node distributions. Since mobility itself requires en-

ergy from its own limited energy source, a deployment scheme

should be designed carefully to minimize energy consumption

during deployment, as well as to achieve certain goals, such as

satisfactory coverage and/or an energy-efficient node topology.

Moreover, it is desirable for a distributed sensor network node to

have a relatively simple hardware architecture, which requires

minimal computing power and memory. Each node should have

1083-4427/$20.00 2005 IEEE


2/15

HEO AND VARSHNEY: ENERGY-EFFICIENT DEPLOYMENT OF INTELLIGENT MOBILE SENSOR NETWORKS 79

a simple and efficient algorithm for deployment, organization,

and management of the network. Even though much research

on energy-efficient organization and management for the static

node topology [23], [30] has been carried out, there has not been

any work on energy efficiency for deployment of mobile nodes

to the best of our knowledge.

Deployment process itself is very energy consuming dueto the locomotive action as well as computation and commu-

nications associated with it. Each node has a limited energy

source. Not only minimizing average moving distance, but also

reducing the difference of the remaining energy among sensor

nodes is essential for a longer system lifetime. Due to the dy-

namic and distributed nature of deployment, it is a challenging

task to obtain full coverage in the ROI and to utilize energy of

each sensor in a relatively fair fashion.

Previous research in distributed-sensor networking has

largely ignored sensor placement issues. Intelligent sensor

deployment strategies are necessary to minimize cost and to

provide sufficient sensor coverage. In addition, sensor de-

ployment must take into account the nature of the terrain,redundancy due to the likelihood of sensor failures, and the

power needed to transmit between deployed sensors.

The deployment of sensor networks varies with the applica-

tion considered. It can be predetermined when the environment

is sufficiently known, in which case, the sensors can be strate-

gically hand placed [2], [19], [22]. Schwiebert et al. [22] re-

strict their investigation to an important class of WSN, namely

biomedical sensor networks, in which the locations of the sen-

sors are fixed and the placement can be predetermined. Bia-

gioni et al. [2] present and analyze a variety of regular de-

ployment topologies, including circular and star deployments as

well as deployments in square, triangular, and hexagonal grids.There exists a close resemblance between the sensor-placement

problem and the traditional art gallery problem(AGP) in com-

putational geometry [19]. The AGP seeks to determine the min-

imal number of positions for guards or cameras so that every

point in a gallery is observed by at least one guard or camera. A

deterministic solution can be found for the AGP and it appears

to be a possible solution to a variety of sensor-placement prob-

lems. Even though there are many solutions to the AGP, all of

them assume the availability of a good model of the environment

a priori. However, it is virtually impossible to have complete in-

formation regarding the environment, where a WSN is likely to

be deployed. Furthermore, too much communication over long

range to obtain global information requires a huge amount of

energy. This is an unaffordable burden on a system with limited

power supply. Thus, deterministic deployment is impractical for

many reasons, such as the harshness of the deployment region

that may be remote, and inhospitable and the increased cost and

latency due to the large number of nodes deployed [23].

The deployment cannot be determined a priori when the en-

vironment is unknown or hostile in which case the sensors may

be air-dropped from an aircraft [6] or deployed by other means,

generally resulting in a random placement [9], [10], [17], [25].

In this paper, the self-deployment of mobile sensor nodes is con-

sidered. This is quite similar to problems considered in cooper-

ative mobile robotics [5]. Mobile sensors are often desirable,since they can patrol a wide area, and can be repositioned for

better surveillance [21]. Some researchers have considered the

use of mobile robots in sensor networks. A recent work on mo-

bile sensor networks [10] presents a distributed and scalable po-

tential field-based approach for the deployment of mobile sen-

sors. The fields are constructed such that each sensor is repelled

by both obstacles and by other sensors, thereby forcing the net-

work to spread itself through the environment. Winfield [25]considered autonomous dispersion of mobile nodes in a scenario

where mobility is required to cover the entire region due to a

lack of wireless-network connectivity. He used a random diffu-

sion method for node deployment while collecting data over a

fixed surveillance region. In the incremental deployment algo-

rithm [9], nodes are added one at a time. The goal is to max-

imize network coverage under the constraint that nodes main-

tain line-of-sight with each other. Loo et al. considered a system

consisting of a number of cooperating mobile nodes that move

toward a set of prioritized destinations under sensing and com-

munication constraints [16]. They show how individual agents

know when cooperation between agents improves the perfor-

mance and when they should suspend cooperation.A related problem to deployment in WSN is spatial local-

ization [4]. In WSN, nodes need to be able to locate them-

selves in various environments and on different distance scales.

Meguerdichian et al. have considered the problem of location

and deployment of sensors in a WSN from a coverage stand-

point [17]. The problems of coverageand deployment are funda-

mentally interrelated. The authors define the coverage problem

from different points of view, including deterministic, statis-

tical, and the worst and best cases. They implicitly assumed

fixed wireless sensor nodes. They argued that coverage is a pri-

mary performance metric that provides an indication regarding

quality-of-service. They combined computational geometry andgraph theoretic approaches to develop algorithms for coverage

calculations. Coverage in WSN, which is one of the main fo-

cuses in this paper, will be discussed later in Section II. Bulusu

et al.s work [3] is somewhat similar to the deployment problem

that is considered here. They have investigated the problem of

adaptive beacon placement for localization in a WSN. They also

pointed out the lack of viability and inadequacy of fixed and

dense beacon placement in some situations due to node per-

turbation during deployment, noisy environment, and self-in-

terference. By placing additional beacons incrementally, they

achieve empirical adaptation to terrain conditions. Unlike tradi-

tional sensor systems, sensor networks depend on dense sensor

deployment and physical colocation with their targets to accom-

plish their goals. Dense deployment allows the use of redun-

dancy [26], can reduce communication costs [20], and provides

sufficient number of nodes to allow physical colocation.

In this paper, three different deployment methods are pro-

posed. First, a deployment algorithm for mobile nodes is pro-

posed when each node is equally important and a peer-based

structure is obtained. In many WSN scenarios, clustering is em-

ployed to take advantage of local information and to reduce en-

ergy consumption. An intelligent energy-efficient deployment

algorithm for cluster-based WSN is proposed. The key idea of

the second algorithm is the introduction of local clustering [12],

[15] during the deployment process so as to increase the amountof local control over a fraction of the entire ROI. Each node


3/15


decides its own mode to be either in a clustering mode or a

peer-to-peer mode based on its local environment such as the

local density and the remaining energy level in a distributed and

adaptive manner. Finally, an energy-efficient deployment algo-

rithm based on Voronoi diagrams (VDs) is proposed.

The goal of the first method is different from prior work on

the deployment problem. The main objective of the first deploy-ment algorithm is topology improvement for longer system life-

time by utilizing mobility of sensor nodes. A decision and con-

trol mechanism is provided at each sensor during deployment,

rather than random diffusion, which is used in Winfields work

[25]. In contrast to Howard et al. [9], who use an incremental

approach, the nodes in the first algorithm are deployed at the

same time and they organize themselves in an adaptive manner.

Unlike Loo et al. [16], the first algorithm does not require pre-

specified destinations to form an energy-efficient topology.

The significance of the second method is to provide a syn-

ergistic combination of cluster structuring and peer-to-peer

deployment scheme in an intelligent manner in a hostile and

unpredictable environment. The goals of our algorithm are the

realization of the largest possible coverage area of the network,

the formation of an energy-efficient node topology for a longer

system lifetime, and the organization of a hierarchical structure

for easier management and scalability that supports collabora-

tion among nodes. These goals can be achieved by an adaptive

combination of two modes: clustering and peer-to-peer. In a

peer-to-peer mode, each node moves itself to a sparse region

so that the coverage of the network may increase and/or an

energy-efficient node topology may be achieved. In a clustering

mode, each node follows the decision of the cluster-head so that

each node spends its energy in a balanced way and performs

collaborative missions if necessary.The significance of the third method is to provide an estimate

of the lifetime of each node in a distributed fashion by using

local VDs. Each node can determine how long it can survive

and which action is more useful to its longevity for the current

node topology during deployment. The best energy-utilization

point is obtained by comparing utility gains for movement to

different possible node locations.

All three methods in this paper are based on the same as-

sumptions shown in Section II but have different strengths for

possible applications. In practice, sensor capabilities may vary

depending on the requirements for a certain task and the avail-

able budget. The first method can achieve a quick deployment

with simple sensors. The second method can establish clustering

structure during deployment. The third method requires more

computation, but shows local assessment of the performance

and high energy efficiency in mobility. The proper deployment

method can be chosen based on the requirements of the appli-

cations and resources available.

Our deployment algorithms will be more useful in situations

where it is hard to ensure precise initial deployment due to the

fact that the deployment area is too dangerous or inaccessible to

humans. One can envisage an application involving a hazardous

region, where sensors mounted on mobile robots are deployed

from an airborne vehicle [6]. These mobile robots then organize

themselves using algorithms presented in this paper. Randomlyscattered sensors over a battlefield or a hazardous site are not

Fig. 1. Sensor coverage models. (a) Binary sensor and (b) stochastic sensor

models.

likely to form a uniform distribution and provide desired cov-

erage. Modification of WSN topology in an autonomous and

distributed manner using our algorithms can help in improving

coverage and also to prolong expected system lifetime. This is

essential in time-critical applications. For example, if an area is

contaminated by some hazardous material, a properly deployed

sensor network can quickly sense and measure the amount of

hazardous material such as poisonous gas or nuclear leakage.

By fully covering the entire area of interest, the overall condi-

tion can be assessed quickly and this information can be used

for search and rescue missions, as well as for evacuation-routeplanning. In some applications, initial sensor distributions may

be concentrated at specific points, such as elevators and stairs in

a building.

In the next section, the sensor deployment problem is formu-

lated. In Section III, performance metrics for a mobile WSN

are discussed. The deployment algorithms are presented in

Section IV followed by simulation results in Section V. Some

concluding remarks are provided in Section VI.

II. MOBILE-NODE DEPLOYMENT PROBLEM

In a WSN, physical placement or deployment of sensor nodes

is needed prior to the initialization of a network for data acqui-sition and transmission using sensor nodes. The assumptions

made in this paper are described and the deployment problem

is formulated in this section.

It is assumed that all sensor nodes have identical capabil-

ities for sensing, communication, computation, and mobility.

Sensing coverage and communication coverage of each node is

assumed to be ideal, which means that both coverage areas have

a circular shape without any irregularity.

The coverage of each sensor can be defined either by a binary

sensor model [23] or a stochastic sensor model [28] as shown in

Fig. 1. In the binary sensor model, the probability of detection

of the event of interest is one within the sensing range (sR),otherwise, the probability is zero. So the coverage of a sensor

network using the binary sensor model is determined by finding

the union of the areas defined by the location of each sensor

and its sR. In the stochastic sensor model, the probability of

detection of the event of interest follows a decaying function of

distance from the sensor. In this paper, the binary sensor model

is employed.

Computation capability is required at each node to support

a distributed algorithm that includes a reasoning and optimiza-

tion process for deployment and routing. It is assumed that the

initial deployment is random and a distributed-deployment al-

gorithm is executed starting from the initial random topology,

using each nodes mobility. Another assumption is that everynode has the ability to know its own location by some method,


4/15


such as the global positioning system or iterative multilatera-

tion [24]. This locationing ability is needed by each node, while

making a decision regarding its next movement in the deploy-

ment process. Also, it is assumed that there are no errors during

transmission of data and in the calculation of locations. It is fur-

ther assumed that each node has only local information from the

neighboring nodes within its direct communication range (cR).The cR of each node is defined by the maximum distance at

which the signal-to-noise ratio is above a given threshold.

Without loss of generality, the deployment problem for a rect-

angular ROI with a certain number of nodes that form an ad hoc

wireless network is considered. Each node has a limited amount

of energy. The goal is to find the positions and movements of

nodes to achieve maximum coverage and to form a uniformly

distributed wireless network in minimum time and with min-

imum energy consumption. A suite of heuristic algorithms are

developed for this problem and their performances are evaluated

in terms of the performance metrics: coverage, uniformity, and

the time and distance traveled until convergence. These metrics

are described next.

III. PERFORMANCE METRICS IN MOBILE WSN

The selection of suitable measures to compare performances

of different approaches and resulting solutions is an important

issue in a mobile WSN. Coverage, uniformity, and time and

distance traveled prior to convergence are considered as per-

formance metrics in mobile WSN here. Coverage and unifor-

mity are related to the performance of sensor networks after the

deployment of sensors is complete. Time and distance traveled

prior to convergence are directly related to the performance of

the deployment scheme itself.

A. Coverage

Generally, coverage can be considered as the measure of

quality of service of a sensor network. The concept of coverage

as a paradigm for the system-level functionality of multirobot

systems was introduced by Gage [7], [31].

In this paper, coverage [9] is defined as the ratio of the union

of areas (in square meters) covered by each node and the area

(in square meters) of the entire ROI. Here, the covered area of

each node is defined as the circular area within its sensing radius

. Perfect detection of all interesting events in the covered area

is assumed

where

is the area covered by the th node;

is the total number of nodes;

stands for the area of the ROI.

If a node is located well inside the ROI, its complete coverage

area will lie within the ROI. In this case, the full area of that

circle, i.e., , is included in the covered region. If a node

is located near the boundary of the ROI, then only the part of

the ROI covered by that node is included in the computation.Because of the areas covered by nodes that fall out of the ROI

and the overlap of covered areas between nodes, one needs to

use more nodes than simply the ratio of and the area sensed

by a single node.

The overall coverage of a sensor network is composed of the

covered regions of each sensor node. Though the coverage of

a sensor is expressed by a sensor model which is binary or sto-

chastic, the overall coverage of a sensor network depends on thelocations of the sensor nodes in the sensor field. The topology

including the locations and spacing of sensor nodes determines

the overall coverage of the network as well as the expected life-

time of the network.

sR and cR of a node are distinguished in the paper. In gen-

eral, they will be different and accordingly sensing coverage and

communication coverage will be different. Sensing coverage

can be accrued when sensor nodes are connected via wireless

links. If the network is separated by any reason, the area cov-

ered by the subnets that do not have wireless links to the sink

node is lost.

B. Uniformity

Uniformly distributed-sensor nodes spend energy more

evenly through the WSN than sensor nodes with an irregular

topology. When the distances between nodes become similar,

each node can utilize its resources efficiently with the minimum

use of its power and an increased throughput, due to reduction

of the interferences between nodes. So, uniformity of network

topology can be used as a good estimator for the expected

system lifetime. Also, fewer nodes are required to cover an ROI

when nodes are more evenly distributed.

Uniformity can be defined as the average local standard de-

viation of the distances between nodes

where

is the total number of nodes;

is the number of neighbors of the th node;

is the distance between th and th nodes;is the mean of internodal distances between the th

node and its neighbors.

In the calculation of local uniformity at the th node, only

neighboring nodes that reside within its cR are considered.

The uniformity measure is a local measure and is computed

locally because each node has access to local information only.

A smaller value of means that nodes are more uniformly

distributed in the ROI. In uniformly distributed networks,

internodal distances are almost the same; the expected energy

consumption per communication as well as the expected life-

time of each node is almost the same if the nodes were identical

and have the same amount of energy initially. Therefore, it is

expected to have full energy utilization at each node and longersystem lifetime for uniformly distributed networks.


5/15


C. Time

The time spent for deployment [9] is also important in many

time-critical applications, such as search-and-rescue and dis-

aster recovery operations. Mostly, the required time depends

on the complexity of the reasoning and optimization algorithm

and physical time for the movement of nodes. The total time

elapsed is defined here as the time elapsed until all the nodesreach their final locations. This paper focuses on the time spent

for deployment itself and not on data-transmission delays from

a source node to a destination node that is commonly used for

network-performance evaluation and its quality of service.

D. Distance

The average distance traveled [9] by each node is related to

the energy required for its movement. So, the expected distance

traveled is important for the estimation of energy (fuel) required

when each node has a limited energy supply. The variance of the

distance traveled is also important to determine the fairness of

the deployment algorithm and for system energy utilization. Ifthe variance of distance traveled is large, the variance of energy

remaining also is large. The nodes that have less energy than

other nodes exhaust their energy early. Early dead nodes result

in a loss of coverage and the remaining nodes may require an

increased transmission range or a longer routing path.

IV. ALGORITHMS

In this paper, three different deployment methods are pre-

sented. The first method operates in a peer-to-peer mode, where

each node is considered to be equal. The second method is a syn-

ergistic combination of the peer-to-peer method with a cluster-

based method. Clustering, a hierarchical networking concept, is

employed in many WSN scenarios to take advantage of local

information and to reduce energy consumption. Finally, an en-

ergy-efficient deployment algorithm based on VDs is proposed.

A. Distributed Self-Spreading Algorithm (DSSA)

The peer to peer algorithm which is called the DSSA is in-

spired by the equilibrium of molecules, which minimizesmolec-

ular electronic energy and internuclear repulsion. Each particle

determines its own lowest energy point in a distributed manner

and its resulting spacing from the other particles is almost the

same. While deploying a WSN using mobile nodes, one ob-serves that one has an analogous problem. If sensors are located

too close to each other, the gain in coverage from additional sen-

sors is not high. On the contrary, if sensors are located too far

from each other, the coverage regions may not overlap and may

cause a partitioning of the network. Both situations are similar to

internuclear repulsion and attractions between molecules. Op-

timal spacing between sensors in the sense of coverage can be

found by a process similar to the equilibrium of molecules.

To begin with, a specified number of nodes are deployed ran-

domly in a given region, for instance, inside a rectangle. The

sR and cR are assumed to be given. Each node can sense or de-

tect an event within its sR and any pair of nodes within their

cR can communicate with each other. This communication isneeded for finding neighborhoods, obtaining locations of nodes

Fig. 2. Pseudocode for the DSSA.

in the neighborhood, and transmitting and forwarding sensed

data. The neighborhood of a node is defined here as nodes within

its cR. The pseudocode of the algorithm is given in Fig. 2. This

distributed algorithm is executed at each node . The algorithmcontains four parts.


6/15


1) Initialization: In the initialization part, the values of cR,

sR, and the initial node locations are specified. The cR and

sR are assumed to be given. Initial node locations are spec-

ified in terms of a vector that contains the longitude component

and the latitude component of each node location, and is as-

sumed to follow a random distribution. Extension to higher di-

mensions is possible by adding more components in the positionvector. A quantity called expected density, which is a rough es-

timate of the desired density, is required in the algorithm. This

can be calculated by using cR cR , where

is the number of nodes, cR is the cR of each node, and is the

area of the ROI. Thus, expected density is the average number

of nodes required to cover the entire area when these nodes are

deployed uniformly. Initial local density of a node is equal

to the number of nodes within its cR. These densities will be

used when decisions regarding positions of nodes are made.

2) Partial Force Calculation: The concept of force is intro-

duced to define the movement of nodes during the deployment

process. The force is dependent on not only the distance be-

tween the nodes but also the current local density. The forcecorresponding to high local density is greater than the force cor-

responding to low local density. The force from a node that is

closer is greater than that from a node that is farther, which is

similar to the movements of the particles in physics that follow

Coulombs Law.

A force function is defined which satisfies the following

conditions.

i) Inverse relation: , when , where

and are node separations from the origin. The node

under consideration is assumed to be at the origin.

ii) Upper bound: .

iii) Lower bound: , where cR, is the nodeseparation and cR is the communication range of each

node.

Condition (i) is the same as in physics, but conditions (ii)

and (iii) are included to modify the model to incorporate the

notion of locality. In other words, a limiting function is applied

via conditions (ii) and (iii).

The partial force at time step on the th node from the th

node that is in the neighborhood of the th node is calculated to

be a repulsive force as

cR (1)

where

stands for the location of the th node at time step ;

stands for the local density of the th node at time

step .

The density factor , which is defined as the ratio of

the local density and the square of the expected density

at each node, is small in sparse regions and is large in dense re-

gions. Its inclusion in the force function expedites the process

of node spreading. Also, internodal distance affects the par-

tial force inversely. Closely located nodes impose larger partial

forces and nodes that are far apart induce smaller partial forceson each other. The magnitude of the partial force exerted by a

Fig. 3. Illustration of nodal movement.

Fig. 4. Illustration of nodal movement in the presence of a failed node.

pair of nodes on each other are the same with the only difference

that the directions are opposite to each other.

After adding all the partial forces at the current node location,

each node decides its next movement. This process provides

a local decision, which includes the consideration of its local

situation, such as the locations of the neighboring nodes anddead node(s), if any. The local information is collected from the

nodes that are within the cR and that information is used for

the calculation of the local density at each node. Each nodes

movement is decided by the combined force at that node due to

nodes in its neighborhood.

3) Oscillation Check: An important issue is determining

when a node should stop its movement. Two stopping criteria

are introduced in the DSSA. If a node moves back and forth

between almost the same locations many times, this node is

regarded to be in the oscillation state. By examining the history

of its movement, each node can determine if oscillations are

going on. One counts the number of oscillations and if thisoscillation count is over the oscillation limit ,

the movement of that node is stopped at the center of gravity

of the oscillating points.

4) Stability Check: If a node moves less than threshold for

the time duration , this node can be con-

sidered to have reached the stable status and that node stops its

movement. This stopping criterion is useful for stationary nodes

because of either exhausted fuel or broken mobile units and also

for the nodes that have reached the stable status.

To illustrate the algorithm, examples of nodal movements are

shown in Figs. 3 and 4.

In Fig. 3, the next movement of node A is considered. Nodes

BE are its neighboring nodes within its cR. The partial forces, and from nodes BE on A are calculated by (1).


7/15


Then the total (resultant) force F on A can be obtained by just

adding partial forces on A, i.e., . Note

that this addition includes both magnitude and direction, i.e., it

is a vector addition. Node A will move to the new position A

as shown. Node C is the closest node to A among neighboring

nodes, so their spacing should be increased. As one can see,

partial force dominates the total force F and A moves awayfrom C.

When sensor nodes are deployed in a remote and hostile re-

gion, some nodes can be adversely affected during and after de-

ployment. Some nodes can lose their mobility and other nodes

can lose their communication functionality. So, a robust deploy-

ment method that can overcome these situations is needed. The

first algorithm exhibits this kind of robustness. First, when a

sensor node loses its mobility, that sensor node does not move

and is considered to be an early stopped node. However, this

node can still be used as a static node in the sensor network.

Neighboring nodes, if they can move, may still improve the ir-

regular topology. Second, when a sensor node loses its com-

munication capability, that sensor node is of no use in a sensor

network. In that case, neighboring nodes may move to the loca-

tions so that the uncovered region can be covered by them. This

case is illustrated in Fig. 4.

The node positions are the same as in the previous example.

Suppose node C is broken during the random deployment pe-

riod and it cannot use its communication unit. Then, neigh-

boring nodes including node A and disregard this broken node

during the calculations of their next movements. The largest

partial force in the previous example is now excluded, i.e.,

. The next position for node A is A and

the lost coverage from the loss of node C can be recovered by

node A. Since the movement of each node is only affected by thecurrent status of neighboring nodes, each node adapts to envi-

ronment changes, such as node failures, various terrain shapes,

etc., and changes its position in an autonomous manner to max-

imize coverage and uniformity.

B. Intelligent Deployment and Clustering Algorithm (IDCA)

In many WSN scenarios, clustering is employed to take ad-

vantage of local information and to reduce energy consumption.

By introduction of local clustering [12], [15] during the deploy-

ment process, it is possible to improve the energy-consumption

characteristics of sensor nodes. Each node decides its own modeto be either in a clustering or peer-to-peer mode based on its

local environment, such as the local density and the remaining

energy level in a distributed and adaptive manner. We call this

algorithm the IDCA. The pseudocode of the algorithm is given

in Fig. 5. This distributed algorithm is executed at each node .

The IDCA also contains four parts, like the DSSA.

1) Initialization: The same as the DSSA algorithm.

2) Mode Determination and Partial Force Calculation: In-

tuitively, sensor nodes in a dense region need to move to a sparse

region to improve coverage and connectivity of a sensor net-

work. Node movement and also the corresponding energy con-

sumption is expected in both sparse and dense regions. Nodes

in a region with the correct node density do not need to moveand spend their energy to improve the performance in terms of Fig. 5. Pseudocode for the IDCA.


8/15


uniformity. The reason is that frequent movements of neigh-

boring nodes may degrade existing uniformity and the energy

spent to improve uniformity is simply wasted. By delaying the

movement of sensor nodes in a region with correct node density

until nodal movements stabilize to some extent, inefficient en-

ergy usage of those sensor nodes can be improved.

Based on the relation between the local density ( ) and theexpected density , the mode at a node is determined. If is

close to the expected density , the node selects the clustering

mode. Nodes in regions that have desired density levels are not

expected to move much to improve coverage and/or uniformity

of sensor nodes. A sensor node in such regions determines its

movement based on its remaining energy level relative to its

neighbors. To begin with, partial force in the clustering mode

is calculated by using (1) as used in the DSSA algorithm. Then,

this partial force is modified by its rank based on its energy level

in the neighborhood. The remaining energies of neighboring

nodes are rank ordered. If a sensor node has the rank among

nodes in the neighborhood, the energy factor is and the

partial force calculated by (1) is multiplied by this factor. If theremaining energy level is low, the partial force of the node will

be smaller than that used in DSSA based on its energy factor

in its neighborhood. The node in this situation saves its energy

and contributes less to WSN performance improvement. If the

remaining energy level is relatively high among the nodes in the

neighborhood, the partial force is determined according to its

rank in its neighborhood. The node in this situation uses its en-

ergy more and contributes more for performance improvement

of the WSN. This energy consideration in the clustering mode

reduces the variation of the remaining energy among sensor

nodes. If of a sensor node at any time is different than at

the current location, this node selects the peer-to-peer mode andpartial force calculation is done by using (1).

3) Oscillation and Stability Checks: Same as the DSSA al-

gorithm. In this paper, the same stopping criteria are used for

both modes: peer to peer mode and clustering mode. However,

IDCA algorithm may use different local metrics and stopping

criteria in a clustering mode.

C. VD-Based Deployment Algorithm (VDDA)

Many researchers have demonstrated the importance and use-

fulness of VDs in various fields, such as mathematics, compu-

tational geometry, biology, chemistry, geography, communica-

tions, and coding theory [1], [29]. Given some number of pointscalled generators or sites in the ROI, their corresponding VD

divides the region according to the nearest-neighbor rule. Each

given point (a site or generator) is associated with the subre-

gion consisting of any points in the ROI that are closest to it. If

these points (generators or sites) are assumed to be sensor loca-

tions and the subregions defined by the corresponding Voronoi

regions are covered by the sensor in the subregion, one has a

possible solution to the deployment problem. A deployment al-

gorithm based on the notion of Voronoi regions is developed. In

this paper, the goal is to have the Voronoi region corresponding

to a sensor to be coincident with the coverage area defined by

the sensor model. The Voronoi region corresponding to a sensor

at a certain time instance is considered as the desired solutionin terms of coverage. At the same time, each sensor has its

coverage defined by the sensor model. If there is any discrep-

ancy between the current sensor coverage and the corresponding

Voronoi region, action needs to be taken to align the two by

sensor movement and resulting changes in topology. When the

two are aligned within predefined tolerance, nodal movements

will be halted and the resulting solution will be accepted. We

call this the VDDA. The pseudocode of the algorithm is givenin Fig. 6. This distributed algorithm is executed at each node .

1) Initialization: In this part, the values of the cR, the sR,

and the initial node locations are specified. Initial energy

for the th node is specified and each node is assumed to

have the same amount of energy in the beginning. Energy con-

sumption for different activities is specified. Energy consump-

tion for movement of the th node is specified in terms of

defined as the cost for movement per unit distance. The total

cost for movement of the th node is equal to the product of

and the distance. Energy consumption per unit time for com-

munication by the th node is denoted by and is a function

of the largest distance between itself and its neighbors. Energy

consumption per unit time for sensing and computation by theth node is assumed to have a fixed cost.

In order to determine the degree to which each node achieves

coverage in terms of energy efficiency, a node utility metric is

considered. The nodeutility metric of the thnodeat time

step is defined as

where

represents the effective area covered by the th node at

time step ;

represents the estimated lifetime of the th node at time

step .

The node utility metric indicates how well the node is utilized

to sense over its effective area during its lifetime.

The local VD is used to calculate the effective area of each

node. First, the initial local VD is obtained using and

cR. For all the points in the Voronoi region corresponding to

the th node , the nearest sensor node is the th node. The

sensing area of the th node is the circular area centered at

with the radius sR. Then the initial effective area is obtained

by the intersection of the Voronoi region of the th node and the

sensing area of the th node. The effective area of the th node

means the area covered by the th node based on the nearest

node concept. Each node has only local information from theneighboring nodes. The node can calculate its VD with the in-

formation of the neighbors. When a different cR is used, the size

of the neighborhood is different and the VD is also different.

The estimated lifetime of th node at time step is de-

fined as

where

stands for the energy remaining at the th node after

time step ;

stands for the energy consumption per unit distancefor movement of the th node at time step ;


9/15


Fig. 6. Pseudocode for the VDDA.

stands for the distance moved by the th node be-

tween time step and time step ;

stands for the energy consumption per unit time due

to communication by the th node;

stands for the energy consumption per unit time for

sensing and computation by the th node.

The node utility metric is used to decide energy-efficientmovements by each node.

2) Finding the Best Energy-Utilization Point: In this part,

the best energy-utilization point of each node is obtained by

comparing utility gains for potential movement to different pos-

sible node locations. Moving from the current location to dif-

ferent node locations will incur a different energy cost for move-

ment and the resulting node topology may or may not reduce

the communication cost. Because the search space is infinite

when continuous coordinates are considered, the reduction of

search space is necessary to some extent. In this paper, local

VDs are used to reduce the search space. Due to the nearest

neighbor relation of VDs, the Voronoi region of the th node will

be more likely covered by the th node than any other nodes. Inthis sense, the local Voronoi region can be considered as the es-

timated or desired coverage by the local node. Moving to the

centroid of the Voronoi region can be beneficial in terms of

coverage and/or uniformity. The centroid of the Voronoi region

is obtained by using the Matlab function ployarea. Also, the

center of the Voronoi range, which is defined as the midpoint

of maximum and minimum along the and coordinates in

the Voronoi region, can be used to guide the search. The search

space is reduced to several points linearly spaced, starting from

the current location to the centroid of the Voronoi region, and

from there, to the center of the Voronoi range. For these points,

the node utility metric is evaluated and the best action is de-termined. Because the best solution is kept during the search

process, a locally optimal solution is obtained after the search

process. After finding the best energy-utilization point for each

node, actual movement of each node occurs at a single time to

save energy.

V. EXPERIMENTAL RESULTS

The performance of the heuristic algorithms in the paper is

evaluated by simulation. In the experiment, 30 randomly placed

nodes in a region of size 10 10 are used to run the DSSA, the

IDCA, and the VDDA. The sR and cR used in the experiment

are 2 and 4 m, respectively.

In Fig. 7, the node locations and coverage of the initial

random deployment before running the algorithms are shown.

Tiny circles represent the positions of nodes and small (shaded)

and large circles are used to show the sR and cR of the nodes,

respectively. Sensor information may be collected within the sR

and communications between nodes are possible within the cR.

Communications are possible between nodes that are connected

by a line in the figure. As seen in Fig. 7, some parts of the re-

gion cannot be covered by the randomly dispersed nodes, even

though there are sufficient nodes in the given ROI. The coverage

is obtained by adding up small areas that are considered in the

sR of any nodes, not considering the connectivity between theareas. It is possible that the areas are not connected. In that


10/15


Fig. 7. Initial distribution of sensor nodes.

Fig. 8. Final node distribution after running DSSA.

case, the actual coverage that can be reported to the user is the

area that is connected to the sink node, not the arithmetic sum

of disconnected areas. In this particular example, the network

is not fully connected, so the actual coverage is much smaller

than just adding the coverages of the disconnected covered

regions. Similar situations can occur when random deployment

is employed, regardless of the number of sensors used for

deployment. The calculated coverage is more than 85% in

Fig. 7, but the actual coverage is well below 50%, because the

network is partitioned in two parts. This situation is exactly the

case where topology improvement is required.

Fig. 8 shows the node locations and coverage after running

the DSSA. The ROI is fully covered after running the algo-rithm. The parameter values used in this simulation run are:

Fig. 9. Sensor-node movements when DSSA is applied.

stable status limit 5, oscillation limit 5, and

threshold for oscillation and stable status 0.1522. Now, the

network is fully connected and also covers the entire ROI. Note

that the spatial node distribution is more uniform than the initial

random distribution shown in Fig. 7.

Fig. 9 shows the actual paths of individual nodes as they

moved from their initial locations to their final locations using

DSSA. Blank circles represent the initial locations and filled cir-

cles indicate the final locations. For the initial distribution of

Fig. 7, each node moves a distance of 3.8485 on average andthe standard deviation of distance traveled is 1.6148. When the

average distance traveled is small, the corresponding energy for

locomotion is small. Also, when the standard deviation of dis-

tance traveled is small, the variation in energy remaining at each

node is not significant and a longer system lifetime with full cov-

erage can be expected.

For the purpose of comparison, a simulated annealing-based

algorithm (SABA) was also used for topology improvement.

The SABA is known as a good solution of many combinato-

rial optimization problems. To implement a SABA for topology

improvement, four main design issues need to be considered.

These are: the definition of the neighborhood, move operator,local energy calculation, and annealing schedule. The definition

of neighborhood used here is the same as for DSSA, i.e., it is

set equal to the cR. This concept of neighborhood is reasonable

because each node can only reach the neighboring node using

single hop communication. The move operator is chosen to be a

random movement within the neighborhood. Local energy cal-

culation is done by adding up the subforces in the neighbor-

hood just like in our DSSA. An exponential cooling schedule

is used as the annealing schedule for efficiency as in [13]. In

SABA, if the energy of the proposed solution is

less than that of the old solution , the proposed solution

is accepted as the new solution. Otherwise, the proposed solu-

tion is accepted with a certain probability , which is given by, where isthe energy and


11/15


Fig. 10. Final node distribution after running SABA.

Fig. 11. Node movements when SABA is applied.

is the current temperature. The parameters used are initial tem-

perature and the stopping criterion is three consecutive

failures in achieving the desired acceptance ratio, defined here

as the frequency of acceptance to the number of trials. The SA

used here is a modified version due to the considerations of timeand energy constraints to mobile node deployment as well as

the distributed nature of the algorithm. The SABA used in this

paper does not necessarily obtain the globally optimal solution.

The result after applying SABA is shown in Fig. 10.

Fig. 10 shows that SABA also works well for the initial distri-

bution shown in Fig. 7. The entire area is covered by 30 sensor

nodes and these nodes are well spread over the region. Fig. 11

shows how individual nodes move from initial locations to final

locations in SABA. For the initial distribution of Fig. 7, each

node moved a distance of 46.4697 on an average and the stan-

dard deviation of distance traveled is 14.5264. Compared with

DSSA, SABA involves more travel distance on an average until

convergence and the corresponding energy required is muchgreater than that of DSSA. Because the standard deviation of

Fig. 12. Final node distribution after running IDCA.

Fig. 13. Sensor node movements when IDCA is applied.

travel distance is also large, the system lifetime with full cov-

erage attained by SABA is expected to be shorter than DSSA.

The result after applying the IDCA is shown in Fig. 12.

Fig. 12 shows that IDCA also works well for the initial distri-

bution shown in Fig. 7. The entire area is covered by 30 sensor

nodes and these nodes are well spread over the region.

Fig. 13 shows how individual nodes move from their initiallocations to final locations in IDCA. For the initial distribution

of Fig. 7, each node moved a distance of 1.866 on average and

the standard deviation of distance traveled is 0.984 09. Com-

pared with DSSA, IDCA involves less travel distance on av-

erage until convergence and the corresponding energy required

is much less than that of DSSA. Note that the path lengths

between starting positions and ending positions in Fig. 13 are

shorter than those in Fig. 9. Because the standard deviation of

travel distance is also small, the system lifetime with full cov-

erage attained by IDCA is expected to be longer than DSSA.

The result after applying the VDDA can be seen in Fig. 14,

which shows that VDDA also works well for the initial distri-

bution shown in Fig. 7. The entire area is covered by 30 sensornodes and these nodes are well spread over the region.


12/15


Fig. 14. Final node distribution after running VDDA.

Fig. 15. Sensor node movements when VDDA is applied.

Fig. 15 shows how individual nodes move from their initial

locations to final locations in VDDA. For the initial distribution

of Fig. 7, each node moved a distance of 1.5498 on an average

and the standard deviation of distance traveled is 0.671 87. Com-

pared with DSSA and IDCA, VDDA involves less travel dis-

tance on average until convergence and the corresponding en-ergy required is much less than those of DSSA and IDCA. Note

that the path lengths between starting positions and ending posi-

tions in Fig. 15 are shorter than those in Figs. 9 and 13. Because

the standard deviation of travel distance is also small, the system

lifetime with full coverage attained by VDDA is expected to be

longer than DSSA and IDCA.

Next, the performances of DSSA, SABA, IDCA, and VDDA

are evaluated in terms of the metrics presented in Section III.

Coverage, uniformity, time, and distance until convergence for

different algorithms are compared here. Results are presented

in Figs. 1619. These results are obtained for different number

of nodes dispersed over a fixed ROI of size 10 10, i.e., for

different node densities to examine the relation between nodedensities and the performance metrics. The number of nodes

Fig. 16. Coverage versus network size.

Fig. 17. Uniformity versus network size.

varies from 20 to 50 and results are averaged over 100 runs

(initial random distributions) for each node density.

Fig. 16 shows the improvement in coverage area from the ini-tial random deployment for SABA, DSSA, IDCA, and VDDA.

All four algorithms exhibit a similar performance over different

network sizes. The coverage achieved by all the algorithms in-

creases as the network size goes up. As the number of nodes in-

creases, the improvement in coverage diminishes. Even though

the average coverage of random dispersion reaches about 99%

at high node density and this number may appear satisfactory

for many application requirements, random deployment may

not guarantee the intended goal of all the applications. More-

over, even if random deployment can cover 99% of the ROI,

there is a possibility of improvement in the uniformity of in-

ternodal distance to improve the lifetime of a sensor network.

As indicated earlier, the standard deviation of internodal dis-tances is employed as the metric for uniformity of the networks.


13/15


Fig. 18. Termination time versus network size.

Fig. 19. Distance traveled versus network size.

Fig. 17 shows the reduction in the standard deviation from the

initial random deployment case. DSSA, IDCA, and VDDA ob-

tain better uniformity than the initial one and DSSA outperformsIDCA slightly. Though SABA also obtains better uniformity

than the initial one, DSSA, IDCA, and VDDA still outperform

it. The improvement in uniformity saturates as network density

increases.

Fig. 18 shows that IDCA leads to faster deployment than the

DSSA at high node densities on average. Termination time is

measured in the number of iterations until the algorithms stop.

Both DSSA and IDCA outperform SABA on average. Also, the

variation in termination times of DSSA and IDCA is less than

that of SABA over a wide range of number of nodes. This means

that both algorithms are less sensitive to the number of nodes,

i.e., network density in terms of termination time for deploy-

ment than SABA. VDDA takes much longer to terminate thanthe other algorithm.

Fig. 20. Distance traveled versus network size (zoomed in).

Fig. 19 shows the mean distance traveled to reach the final

locations for deployment. DSSA and IDCA require much

less travel distance than SABA. The performance of VDDA

is the best among the four algorithms in terms of distance

traveled. This distance is related to the required energy (fuel)

for deployment.

A zoomed-in version of the mean distance traveled for DSSA,

IDCA, and VDDA is shown in Fig. 20. VDDA requires less

mean distance traveled than DSSA and IDCA, so VDDA saves

more energy for the movement of sensors than the other al-

gorithms. Though DSSA and VDDA have similar coverages

as seen in Fig. 16, VDDA needs a longer time to converge asseen in Fig. 18, VDDA requires less energy for the movement

of nodes than DSSA as shown in Fig. 20, and VDDA can ob-

tain more uniformly distributed node topology than DSSA as

seen in Fig. 17. Therefore, VDDA is more energy efficient than

DSSA in general. Also, it is observed that the required energy as

well as the distance traveled at different node densities is almost

constant, especially in IDCA. Thus, the required energy (fuel)

is quite insensitive to network density in IDCA. This can make

the planning of energy consumption during deployment easier

over a wide range of network densities.

As seen in Figs. 1620, 2540 nodes are required to attain

acceptable performance for the problem considered here. Whentoo few nodes are used, it is unable to obtain full coverage over

the ROI. When too many nodes are used, there is not much gain

in coverage improvement because of the diminishing marginal

gain in terms of coverage, though more uniform distribution still

can be obtained. With the number of nodes in this range, the re-

quired time to converge is almost the same for all the algorithms.

Because the variation in time required to converge and the travel

distance is smaller over this range of node densities, it is easier

to estimate the required energy for deployment. Extensive ex-

periments on nonhomogeneous nodes, nonrectangular regions,

realization of nonuniform target distribution, and robustness of

the algorithms are conducted and results are available in [8].

The performance of the algorithms in this paper is not com-pared with those in [10], [25], and [28] because the assump-


14/15


tions are different. In [10], additional equipment is used to de-

tect other nodes and obstacles. Moreover, nodes are initially de-

ployed over a compact region and then are spread out. In [ 25],

it is assumed that there are insufficient robots in a physically

bounded region. In [28], it is assumed that global information

regarding other nodes is available.

Experiments using different values for sR and cR were con-ducted and similar performances are observed. The algorithms

were simulated in environments that are not rectangular and the

performances are similar to those of rectangular regions. These

results are not included in this paper due to length considera-

tions. Results are available in [8].

VI. SUMMARY

The deployment problem for mobile WSN is considered in

this paper. A ROI needs to be covered by a given number of

nodes with limited sensing and cR. A random distribution of

nodes over the ROI is assumed as the initial node distribution.

Though many scenarios adopt random deployment for practical

reasons, such as deployment cost and time, random deploy-

ment may not provide a uniform distribution, which is desir-

able for a longer system lifetime over the ROI. In this paper, a

number of distributed algorithms for the deployment of mobile

nodes are proposed to improve an irregular initial deployment

of nodes. A peer-to-peer algorithm analogous to the equilib-

rium of molecules and an enhanced intelligent energy-efficient

deployment algorithm for cluster-based WSN by a synergistic

combination of cluster structuring and peer-to-peer deployment

scheme are proposed. A distributed algorithm using VDs based

on local computation is also proposed. After employing these

algorithms, the ROI is covered by more uniformly distributednodes. While developing these algorithms, one should consider

factors such as density of nodes, memory constraints, local-

ization errors, and scalability of mobile nodes (network size).

Through mobility and locationing ability of nodes, these algo-

rithms provide a way to avoid expensive redeployment process.

This postdeployment idea is quite useful for many situations, es-

pecially when a large fraction of nodes are destroyed or broken

during deployment or operation in a hostile situation, or where

initial distribution is quite uneven and when human intervention

for redeployment is too costly or too risky. The performance of

these algorithms is determined by the percentage of region cov-

ered, computational/deployment time, the mean distance trav-

eled required for deployment, and uniformity of the networks.

Simulation results show that the proposed algorithms success-

fully obtain a more uniform distribution from initial uneven dis-

tributions in an energy-efficient manner.

In this paper, only one-hop neighbors were included while

making the decision regarding next nodal movement. However,

better solutions in terms of energy efficiencymay be found when

a wider neighborhood is used. Computation cost, time delay,

energy consumption, and tradeoffs between them for different

neighborhood sizes will be the main issues to be considered

in this direction. This will require the inclusion of multihop

neighbors. Also, the relation between cluster-size and neighbor-

hood size will need to be established for energy saving duringdeployment.

In practice, a WSN is deployed over large regions. The ROI

can be divided into multiple sub-ROIs for easy deployment, or-

ganization, and management. Hierarchical deployment schemes

may be investigated to handle the scalability issue in WSNs. The

determination of the size of sub-ROI and their corresponding

density and edge effects due to the division of the ROI are worth

pursuing. The effect of uncertainty in sensor-node locations onthe performance of our algorithms is another area for further

investigation.

REFERENCES

[1] F. Aurenhammer, Voronoi diagramsA survey of a fundamentalgeometric data structure, in ACM Comput. Survey, vol. 23, 1991, pp.345405.

[2] E. Biagioni and G. Sasaki, Wireless sensor placement for reliable andefficient data collection, in Proc. Hawaii Int. Conf. Syst. Sci., Jan. 2003,p. 127b.

[3] N. Bulusu, J. Heidemann, and D. Estrin, Adaptive beacon placement,in Proc. 21st Int. Conf. Distributed Comput. Syst., Apr. 2001, pp.489498.

[4] , GPS-less low-cost outdoor localization for very small devices,

IEEE Pers. Commun., vol. 7, no. 5, pp. 2834, Oct. 2000.[5] Y. U. Cao, A. Fukunaga, and A. Kahng, Cooperative mobile robotics:

Antecedentsand directions,Autonomous Robots,vol.4,pp.123,1997.[6] S. S. Dhillon and K. Chakrabarty, Sensor placement for effective cov-

erage and surveillance in distributed sensor networks, in Proc. IEEEWireless Commun. Netw. Conf., 2003, pp. 16091614.

[7] D. W. Gage, Command control for many-robot systems, presented atthe 19th Annu. AUVS Tech. Symp, Huntsville, AL, June 2224, 1992.

[8] N. Heo, Distributed deployment algorithms for mobile wireless sensornetworks, Ph.D. dissertation, Dept. Elect. Eng. Comput. Sci., SyracuseUniv., Syracuse, NY, 2004.

[9] A.Howard,M. J.Mataric, and G.S. Sukhatme, An incremental self-de-ployment algorithm for mobile sensor networks, Autonomous Robots,vol. 13, no. 2, pp. 113126, 2002.

[10] A. Howard, M. J. Mataric, and G. S. Sukhatme, Mobile sensor networkdeployment using potential fields: A distributed, scalable solution to the

area coverage problem, in Proc. 6th Int. Conf. Distributed AutonomousRobotic Syst., Fukuoka, Japan, 2002, pp. 299308.[11] C. E. Jones, K. M. Sivalingam, P. Agrawal, and J. C. Chen, A survey

of energy efficient network protocols for wireless networks, WirelessNetw., vol. 7, no. 4, pp. 343358, 2001.

[12] V. Kawadia and P. R. Kumar, Power control and clustering in ad hocnetworks, in Proc. IEEE INFOCOM Conf., 2003, pp. 459469.

[13] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, Optimization by simu-lated annealing, Science, vol. 220, pp. 671680, 1983.

[14] S. Kumar, F. Zhao, and D. Shepherd, Collaborative signal and informa-tion processing in microsensor networks, IEEE Signal Process. Mag.,vol. 19, no. 2, pp. 1314, Mar. 2002.

[15] C. R. Lin and M. Gerla, Adaptive clustering for mobile wirelessnetworks, IEEE Journal on Sel. Areas Commun., vol. 15, no. 7, pp.12651275, Sep. 1997.

[16] L. Loo, E. Lin, M. Kam, and P. K. Varshney, Cooperative multi-agentconstellation formation under sensing and communication constraints,

in Cooperative Control and Optimization. Norwell, MA: Kluwer,2002, pp. 143170.

[17] S. Meguerdichian, F. Koushanfar, M. Potkonjak, and M. Srivastava,

Coverage problems in wireless ad hoc sensor networks, in Proc. IEEEINFOCOM Conf., 2001, pp. 13801387.

[18] R. Min et al., Energy-centric enabling technologies for wireless sensornetworks,IEEE Wireless Commun., vol. 9,no.4, pp. 2839, Aug. 2002.

[19] J. ORourke, Art Gallery Theorem and Algorithms. New York, NY:Oxford University Press, 1987.

[20] G. J. Pottie and W. J. Kaiser, Embedding the internet: Wireless inte-gratednetwork sensors, Commun. ACM,vol.43,no.5,pp.5158, 2000.

[21] H. Qi, S. S. Iyengar, and K. Chakrabarty, Distributed sensor fusion-areview of recentresearch,J. Franklin Inst., vol. 338,pp. 655668,2001.

[22] L. Schwiebert, S. K. S. Gupta, and J. Weinmann, Research challengesin wireless networks of biomedical sensors, in Proc. ACM/IEEE Conf.

Mobile Comput. Netw., 2001, pp. 151165.

[23] S. Slijepcevic and M. Potkonjak, Power efficient organization of wire-less sensor networks, in Proc. IEEE Int. Conf. Commun., vol. 2, 2001,pp. 472476.


15/15


[24] K. Sohrabi, B. Manriquez, and G. Pottie, Near-ground widebandchannel measurements, in Proc. 49th Veh. Technol. Conf., 1999, pp.571574.

[25] A. F. T. Winfield, Distributed sensing and data collection via brokenad hoc wireless connected networks of mobile robots, in Distributed

Autonomous Robotic Systems 4, L. E. Parker, G. Bekey, and J. Barhen,Eds. New York: Springer-Verlag, 2000, pp. 273282.

[26] Y. Xu, J. Heidemann, and D. Estrin, Geography-informed energy con-

servation for ad hoc routing, in Proc. ACM/IEEE Int. Conf. MobileComput. Netw., July 2001, pp. 7084.[27] W. Ye, J. Heidemann, and D. Estrin, An energy-efficient MAC protocol

for wireless sensor networks, in Proc. IEEE INFOCOM Conf., vol. 3,2002, pp. 15671576.

[28] Y. Zou and K. Chakrabarty, Sensor deployment and target localizationbased on virtual forces, in Proc. IEEE INFOCOM Conf., vol. 2, 2003,pp. 12931303.

[29] Voronoi [Online]. Available: http://www.voronoi.com/[30] J. E. Wieselthier, G. D. Nguyen, and A. Ephremides, Algorithms for

energy-efficient multicasting in static ad hoc wireless networks,MobileNetw. Applicat., vol. 6, no. 3, pp. 251263, 2001.

[31] D. W. Gage, Command control for many-robot systems, UnmannedSyst. Mag., vol. 10, no. 4, pp. 2834, 1992.

Nojeong Heo received the B.S. degree in electricalengineering from Seoul National University, Seoul,Korea, in 1996 and the M.S. and Ph.D. degrees inelectrical engineering from Syracuse University,Syracuse, NY, in 1999 and 2004, respectively.

He is currently a Senior Engineer at SamsungElectronics Co., Ltd. His research interests includewireless communication, ad hoc networks, sensor

networks, and next-generation mobile networks.

Pramod K. Varshney (S72M77SM82F97)was born in Allahabad, India, on July 1, 1952. Hereceived the B.S. degree in electrical engineering andcomputer science (with highest honors) and the M.S.and Ph.D. degrees in electrical engineering from theUniversity of Illinois, Urbana, in 1972, 1974, and1976, respectively.

He is currently Research Director of The New

York State Center for Advanced Technology inComputer Applications and Software Engineering(CASE). During 19721976, he held teaching and

research assistantships at the University of Illinois. Since 1976, he has beenwith the Department of Electrical and Computer Engineering, SyracuseUniversity, Syracuse, NY, where he is currently Professor of electrical engi-neering and computer science. He has served as the Associate Chairman of thedepartment from 1993 to 1996. His current research interests are in distributedsensor networks and data fusion, detection and estimation theory, wirelesscommunications, image processing, remote sensing, radar signal processing,and parallel algorithms. He has supervised 34 Ph.D. dissertations, authored orcoauthored over 85 journal papers and over 250 conference papers. He is theauthor ofDistributed Detection and Data Fusion (New York: Springer-Verlag,1997). He has consulted for General Electric, Hughes, Booz-Allen andHamilton, SCEEE, Kaman Sciences Corp., Andro Computing Solutions, ITT,and Digicomp Research.

Dr. Varshney isa memberof TauBetaPi and isthe recipientof the 1981 ASEE

Dow Outstanding Young Faculty Award. He was elected to the grade of Fellowof the IEEE in 1997 for his contributions in the area of distributed detection anddata fusion. In 2000, he received the Third Millennium Medal from the IEEEand Chancellors Citation for Exceptional Academic Achievement at SyracuseUniversity. He was the Guest Editor of the Special Issue on Data Fusion of thePROCEEDINGS OF THE IEEE, January 1997. Heis on the editorialboard ofClusterComputing Information Fusion. He is a Distinguished Lecturer for the IEEEAES Society. He was the President of the International Society of InformationFusion in 2001. While at the University of Illinois, he was a James Scholar, aBronze Tablet Senior, and a Fellow.

Date post:	14-Apr-2018
Category:	Documents
Upload:	gbismpikis
View:	219 times
Download:	0 times

Coverage Problem 2

Documents